script-doctor

name	script-doctor
description	Use when evaluating scripts for StoryVerse AI production readiness, diagnosing AI-human video risks, creating risk cards, AI-ready scene packages, asset maps, beat tables, storyboard prompt plans, or standardizing Hermes/Slack Script Doctor experiments.

Script Doctor

Overview

Use this skill to turn a raw script, scene, beat table, or production result into a structured Script Doctor diagnosis for AI-human video generation. The goal is not to praise or rewrite by default; the goal is to decide what can be produced, what will fail, why it will fail, and what production preparation is required before StoryVerse generation.

This repository is agent-agnostic. Codex and Claude Code can load SKILL.md as a skill-style instruction file. Hermes, OpenClaw, and Slack bots can copy the core rules plus the experiment SOP into their system prompt or pinned workflow.

Core Workflow

1. Confirm scope

   • Project-level: evaluate a full script or episode batch.
   • Scene-level: evaluate one scene for generation readiness.
   • Prep-level: produce an AI-ready scene package.
   • Experiment-level: standardize Hermes/Slack case testing. Ask only when scope materially changes the output.

2. Classify the case Use the closest primary type, and add secondary tags when needed:

   Type	Trigger	Common recommendation
   VFX / safety / action high-risk	pregnancy visuals, blood, bed scenes, spirits, transformation, falling, crowds	Degrade into sound, reaction shots, POV, props, cutaways
   Subtext / missing direction	political dialogue, power games, hidden meaning, weak blocking	Require director-style shot design or production rewrite
   Visual hook but asset-lock required	clear action premise with car, station, chase, corpse, weapon, costume	Generate asset map first, then split into controllable beats
   Literary beat not executable	elegant prose but missing shot size, camera, action boundary	Translate into shot number + scene + action + expression keywords

3. Score with production meaning Use a 0-100 fit score and explain it plainly. Recommended dimensions:

   • Visual hook strength.
   • Script production completeness.
   • Character / multi-person consistency risk.
   • Action and prop continuity risk.
   • VFX / safety / moderation risk.
   • Asset and costume complexity.
   • Rewrite cost before generation.

4. Create evidence-backed risk cards Every issue must include source evidence, why AI generation struggles, and a concrete next action.

   Issue:
   Severity: Blocker / High / Medium / Low
   Source evidence:
   Why AI struggles:
   Recommended fix:
   Next action:
   

5. Recommend one production path Choose exactly one primary next step:

   • Proceed to pipeline.
   • Generate asset map first.
   • Generate AI-ready scene package.
   • Rewrite scene before storyboard.
   • Do not produce this scene until human direction is added.

6. If asked for AI-ready prep Keep original dialogue and character setup unless the user explicitly asks to change them. Output:

   • rewritten action / blocking details;
   • beat table with durations;
   • asset map with priorities;
   • storyboard keyframe prompt structure;
   • human checkpoints before generation.

Output Rules

• Lead with the decision, not a long essay.
• Separate diagnosis from rewrite. Do not silently rewrite the script inside an evaluation.
• Prefer concrete production language: "三人同框会漂移" beats "可能有挑战".
• Do not pretend multi-character consistency can be fully locked. Say it can only be reduced through asset references, stable costume, fewer simultaneous characters, and smaller shots.
• Do not turn literary or emotional lines directly into prompts. Translate them into visible action, blocking, gaze, prop state, shot size, and edit rhythm.
• For unsafe or hard-to-render content, recommend substitution through sound, aftermath, reaction, POV, cutaway, or symbolic prop.
• When evidence is missing, mark the assumption instead of inventing certainty.

Hermes / Team Experiment Mode

When the user asks how to run tests with business users, Hermes, Slack, case review, or SOP, read references/experiment-sop.md and produce a process that includes:

• fixed input template;
• fixed Script Doctor output schema;
• business reviewer rubric;
• case library fields;
• weekly rule update cadence.

For Chinese business-team usage, use references/experiment-sop.zh-CN.md.

Templates

When the user asks for a reusable report, prompt, case card, or team-facing format, read references/output-templates.md.

For Chinese report templates, use references/output-templates.zh-CN.md.